A Python library for explainability of machine learning algorithms for spectroscopic data in an agnostic, deterministic, and simple way.

• GitHub repo: https://github.com/pudu-py/pudu
• Documentation: https://pudu-py.github.io/pudu
• PyPI: https://pypi.python.org/pypi/pudu
• Conda-forge: https://anaconda.org/conda-forge/pudu
• Free software: MIT license

Introduction

pudu is a Python package that helps interpret and explore the results of machine learning algorithms with spectroscopic data. It does this by quantifying the change in the prediction according to the change in the features. This library works with classification and regression problems with both 1-d and 2-d problems. actiIn order to see the exact procedure and format needed, please refer to the examples in the docs.

Features

The following is a list of the main features that pudu package enables.

• Importance: measures the change in prediction according to perturbations in the features.
• Speed: calculates how fast a prediction changes according to different perturbation levels.
• Synergy: tests the synergy between features and the change in classification probability.
• Re-activations: Evaluates how activations maps from CNN’s change according to perturbations in the data.
• Easy plotting with ample personalization options for all the cases above.

Quickstart

1. Install this library using pip::

    pip install pudu
   

2. Install this library using conda-forge::

    conda install -c conda-forge pudu
   

3. Test it by running one of the examples in the docs.

4. If you find this library useful, please consider a reference or citation as::

    @misc{Grau-Luque2023Pudu,
    author = {E. Grau-Luque, I. Becerril-Romero, A. Perez-Rodriguez, M. Guc, V. Izquierdo-Roca},
    title = {pudu},
    year = {2023},
    publisher = {GitHub},
    journal = {GitHub repository},
    howpublished = {\url{https://github.com/pudu-py/pudu}},
    }
   

5. Stay up-to-date by updating the library using::

   conda update pudu
   pip install --update pudu
   

6. If you encounter problems when updating, try uninstalling and then re-installing::

    pip uninstall pudu
    conda remove pudu
   

Credits

This package was created with Cookiecutter and the giswqs/pypackage project template.